
今天是加入Zoran Mobile一周年的日子...
Sean Song's Weblog
新家搬进去了,忙的也没太多感觉,就是觉得比原来宿舍吵一点,晚上睡不好,所以更累了...
该来的终于来了,几年前就有客户提卡拉OK打分的功能,不知谁本领大,一直压着.最近压不住了,人家真的要了.想要你就说嘛,你不说我怎么知道你要呢?你要我不会不给你,你不要我没理由给你...
变态的是客户要在I96上作.我那可怜的RAM啊,怎么够作这个呢?就像给我一辆单车,让我20分钟从福田骑到蛇口...好在韩国经理说了,你只需要给出一个非随机数的方案,效果如何我去推...窃喜,随机数+3就是非随机数了吧...
Ares and I checked a CDDA S/PDIF channel status bug today. It is about "copy protection" bit in S/PDIF channel status. There are 3 kind of CDDA disks regarding "copy protection":
-- copy free: the disk can be copied, and its children are also copy free. So it is fecund.
-- copy never: the disk can not be copied. So it is eunuch.
-- copy once: the disk can be copied, but its children are "copy never".
I admire Sony and Philips so much that they could create CD-DA system 20 years ago. They really have long period accumulation of audio (digital + analog) expertise. They make good system definitions which are perfect balance between complexity and efficiency, also they are expandable.
Why they can achieve that? I think there's one key reason -- the combination of Sony and Philips is some kind of combination of Oriental and Western culture.
Born in Shandong, the home of Confucius, I myself was greatly influenced by Chinese traditional culture. Thus Sony and Philips connection should be an excellent example for me in order to have good thoughts, successful mental and physical products. Living abroad for several years sounds necessary.
最近工作中遇到的一系列问题让我开始统一考虑一个问题:嵌入式实时音频DSP系统中,软件MIPs的问题。关于MIPs计算搞DSP的人都清楚,但仔细想来,MIPs只是一个宽泛的概念,在不同场合、不同要求的时候其实是有不同算法的:
1. 算法MIPs:纯算法MIPs,这里面又包括算法定义的MIPs(比如ITU标准Vocoder的算法MIPs)和每个厂商实现起来算法的MIPs(比如优化程度不同,精度折中导致的再优化);
2. 系统MIPs:在算法MIPs的基础上,系统实现起来所消耗的MIPs应该比纯算法MIPs要大,比如有些系统(其实大多数系统都是)需要存取外部内存的数据或程序(通常是SDRAM),这里的SDRAM访问有等待时间问题(DMA waiting cycles);再比如实时系统的I/O有时会有各种中断,中断处理程序(ISR)有些是周期性的,有些非周期性的(异步的),但都需要指令周期(cycles)来执行。这些都增加的系统MIPs开销。
所以猛得提起来MIPs这个概念好像很简单,仔细分析分析是有些内容的。打算找个空闲的时候归纳归纳,写个小文章试试。
//现在想的好,怕有空了就睡觉泡妞没有写文章的动力了